# Full text of "national :: appNotes :: AN"

## See other formats

tn National Semiconductor Application Note AN Current Feedback Amplifiers November 1 992 The use of clever circuit arcliitectures and tlie deveiopment of fiigli-speed complementary BJT processes make it possible to achieve monolithic speeds and bandwidths hitherto available only in hybrid form. — Sergio Franco, San Francisco State University 1600 Holloway Avenue, San Francisco, CA 94132 In their effort to approximate the ideal op amp, manufac- turers must not only maximize the open-loop gain and minimize input-referred errors such as offset voltage, bias current, and noise, but must also ensure adequate band- width and settling-time characteristics. Amplifier dynamics are particuularly important in applications like high- speed DAC buffers, subranging ADCs, S/H circuits, ATE pin drivers, and video and IF drivers.^ Being basically voltage-processing devices, op amps are subject to the speed limitations inherent to voltage- mode operation, stemming primarily from the stray cap- acitances of nodes and the cutoff frequencies of transis- tors. Particularly severe is the effect of the stray capac- itances between the input and output nodes of high-gain inverting stages because of the Miller effect which mul- tiplies the stray capacitance by the voltage gain of the stage. On the other hand, it has long been recognized that current manipulation is inherently faster than voltage manipulation. The effect of stray inductances in a circuit is usually less severe than that of its stray capacitances, and BJTs can switch currents much more rapidly than voltages. These technological reasons form the basis of emitter-coupled logic, bipolar DACs, current conveyors, and the high-speed amplifier topology known as current- feedback.^ Fortruecurrent-modeoperation, all nodes in the circuit should ideally be kept at fixed voltages to avoid the slow- down effect by their stray capacitances. However, since the output of the amplifier must be a voltage, some form of high-speed voltage-mode operation must also be provided at some point. This is achieved by employing gain configurations that are inherently immune from the Miller effect, such as the common-collector and the cascode configurations, and by driving the nodes with push-pull stages to rapidly charge/discharge their stray capacitances. To ensure symmetric rise and fall times, the npn and pnp transistors must have comparable characteristics in terms of cutoff frequency ft. Traditionally, monolithic pnp's have been plagued by much poorer performance char- acteristics than their npn counterparts. However, the recent development of truly complementary high-speed processes makes it possible to achieve monolithic speeds that were hitherto available only in hybrid form. The advantages of the current-feedback topology are best appreciated by comparing it against that of the conventional op amp.^" The Conventional Op Amp The conventional op amp consists of a high input- impedance differential stage followed by additional gain stages, the last of which is a low output-impedance stage. As shown in the circuit model of Fig. la, the op amp transfer characteristic is Vo = a(jf)Vd (1) where Vo is the output voltage; Vd = Vp — Vn is the differential input voltage; and a(jf), a complex function of frequency f, is the open-loop gain. Connecting an external network as in Fig. lb creates a feedback path along which a signal in the form of a voltage is derived from the output and applied to the noninverting input. By inspection. Vd Vi Ri Ri + R2 Vo (2) (a) (b) Fig. 1 : Circuit model of thie conventionai op amp, and connec- tion as a noninverting amplifier. Substituting into Eq. (1), collecting, and solving for the ratio Vo/ Vi yields the familiar noninverting amplifier transfer characteristic A(jf) Vi 1 + R2 1 Ri 1 1 + 1/T(if) (3) where A(jf) represents the closed-loop gain, and a(jf) T(if) 1 + R2/R1 (4) represents the loop gain. © 1992 National Semiconductor Corporation Printed in ttie U.S.A. http://www.national.com The designation loop gain stems from the fact that if we break the loop as in Fig. 2a and inject a test signal Vx with Vi suppressed, the circuit will first attenuate Vx to produce Vn = Vx/(1 + R2/R1), and then amplify Vn to produce Vo = — aVn. Hence, the gain experienced by a signal in going around the loop is Vo/Vx = — a/(1 + R2/R1). The negaf/Veof this ratio represents the loop gain, T^ —(Vo/Vx). Hence, Eq. (4). The loop gain gives a measure of how close A is to the ideal value 1 + R2/R1, also called the noise gain of the circuit. By Eq. (3), the larger T, the better. To ensure substantial loop gain over a wide range of closed-loop gains, op amp manufacturers strive to make a as large as possible. Consequently, Vd will assume extremely small values since Vd = Vo/a. In the limit a ^ °° we obtain Vd -* 0, that is, Vn -^ Vp. This forms the basis of the familiar op amp rule: when operated with negative feedback, an op amp will provide whatever output voltage and current are needed to ideally force Vn to follow Vp. Gain(dB) f(dec) Fig. 2: Test circuit to find the ioop gain, and graphical method to determine the ciosed-ioop bandwidth f^. Gain-Bandwidth Tradeoff Large open-loop gains can physically be realized only over a limited frequency range. Past this range, gain rolls off with frequency. Most op amps are designed for a constant rolloff of — 20dB/dec, so that the open-loop response can be expressed as a(jf) 1 i(f/fa (5) where ao represents the dc gain, and fa is the — 3dB frequency of the open-loop response. Both parameters can be found from the data sheets. For example, the 741 op amp has ao — 2 X 10= and U — 5Hz. Substituting Eq. (5) into Eq. (4) and then into Eq. (3), and exploiting the fact that (1 + R2/Ri)ao<1, we obtain (6) A(jf) = 1 + R2/R1 1 + j(f/fA) where fA = f. 1 + R2 /Ri (7) represents the closed-loop bandwidth, and ft = aofa represents the open-loop unity-gain frequency, that is the frequency at which | a | = 1. For instance, the 741 op amp has f, = 2 X 10^ X 5 = 1 MHz. Equation (7) reveals a gain-bandwidth tradeoff. As we raise the R2/R1 ratio to increase the closed-loop gain, we also decrease its bandwidth in the process. Moreover, by Eq. (4), the loop-gain is also decreased, thus leading to a greater closed-loop gain error. The above concepts can also be visualized graphically. Since Eq. (4) implies | T |dB = 20 log | T | = 20 log | a | - 20 log (1 + R2/R1) = I a IdB - (1 + R2/Ri)dB, it fol- lows that the loop gain can be found graphically as the difference between the open-loop gain and the noise gain. This is shown in Fig. 2b. The frequency at which the two curves meet is called the crossover frequency. At this frequency we have T = 1 /— 90° = — j, that is, I A I = (1 -I- R2/R1) / \/2. Thus, the crossover frequency represents the — 3dB frequency of the closed-loop re- sponse, that is, the closed-loop bandwidth fA. We now see that increasing the closed-loop gain shifts the noise-gain curve upward, thus reducing the loop gain, and causes the crosspoint to move up the | a | curve, thus decreasing the closed-loop bandwidth. Clearly, the circuit with the widest bandwidth and the highest loop gain is also the one with the lowest closed- loop gain. This is the voltage follower, for which R2/ Ri = 0, so that A = 1 and fA = ft- Slew- Rate Limiting To fully characterize the dynamic behavior of an op amp, we also need to know its transient response. If an op amp with the response of Eq. (5) is operated as a unity-gain voltage follower and is subjected to a suitably small voltage step, its dynamic behavior will be similar to that of an RC network. Applying an input step AV will cause the output to undergo an exponential transition with magnitude AVo = AV, and with a time-constant T = 1/{277-ft). For the 741 op amp we have r = 1/(27rX 106) = 170 ns. The rate at which the output changes with time is highest at the beginning of the exponential transition, when its value is AVo/r. Increasing the step magnitude increases this initial rate of change, until the latter will saturate at a value called the slew-rate (SR). This effect stems from the limited ability of the internal circuitry to charge/discharge capacitive loads, especially the com- pensation capacitor responsible for open-loop bandwidth fa. To illustrate, refer to the circuit model of Fig. 3, which is typical of many op amps." The input stage is a trans- conductance block consisting of differential pair Qi— Q2 and current mirror Q3— Q4. The remaining stages are lumped together as an integrator block consisting of an inverting amplifier and the compensation capa- citor C. Slew-rate limiting occurs when the trans- conductance stage is driven into saturation, so that all the current available to charge/discharge C is the bias current / of this stage. For example, the 741 op amp has \ = 20fjA and C = 30 pF, so that SR = l/C = 0.67 VZ/ys. The step magnitude corresponding to the onset of slew-rate limiting is such that AV/r = SR, that is, AV = SR X r = (0.67 V//ys) X (1 70 ns) = 1 1 6 mV As long as the step is less than 116 mV, a 741 voltage follower will respond with an exponential transition governed by r== 1 70 ns, whereas for a greater input step the output will slew at a constant rate of 0.67 V///s. http://www.national.com YV Fig. 3: Simplified slew-rate model of a conventional op amp. In many applications tlie dynamic parameter of greatest concern is the settling time, that is, the time it takes for the output to settle and remain within a specified band around its final value, usually for a full-scale output transition. Clearly, slew-rate limiting plays an important role in the settling-time characteristic of the device. The Current-Feedback Amplifier As shown in the circuit model of Fig. 4, the architecture of the current-feedback amplifier (CF amp) differs from the conventional op amp In two respects:^ 1. The Input stage is a unity-gain voitage buffer con- nected across the Inputs of the op amp. Its function Is to force Vn to follow Vp, very much like a conven- tional op amp does via negative feedback. However, because of the low output impedance of this buffer, current can easily flow in or out of the inverting input, though we shall see that in normal operation this current Is extremely small. (a) (b) Fig. 4: Circuit model of the current-feedback amplifier, and connection as a noninverting amplifier. 2. Amplification is provided by a transimpedance amplifier which senses the current delivered by the buffer to the external feedback network, and pro- duces an output voltage Vo such that Vo = Z(jf)ln (8) where z(jf) represents the transimpedance gain of the amplifier, in V/A or Q, and In is the current out of the inverting input. To fully appreciate the inner workings of the CF amp, it is instructive to examine the simplified circuit diagram of Fig. 5a. The input buffer consists of transistors Qi through Q4. While Qi and Q2 form a low output-Impedance push-pull stage, Q3 and Q4 provide Vbe compensation for the push-pull pair, as well as a Darlington function to raise the input impedance. Summing currents at the inverting node yields h — I2 = In, where I1 and I2 are the push-pull transistor currents. Two Wilson current mirrors, consisting of tran- sistors Qg— Q10— Q11 and Q13— Q14— Q15, reflect these currents and recombine them at a common node, whose equivalent capacitance to ground has been designated as C. By mirror action, the current through this capacitance is Ic = h ~ I2, that is Ic In (9) The voltage developed by C In response to this current is then conveyed to the output via a second buffer, consisting of Q5 through Qs. The salient features of the CF amp are summarized in block diagram form in Fig. 5b. When the amplifier loop is closed as in Fig. 4b, and whenever an external signal tries to imbalance the two inputs, the input buffer will begin sourcing (or sinking) an imbalance current L to the external resistances. This imbalance is then conveyed by the Wilson mirrors to capacitor C, causing Vo to swing in the positive (or negative) direction until the original imbalance In is neutralized via the negative feedback loop. Thus, In plays the role of error signal in the system. Fig. 5: Simplified circuit diagram and block diagram of a current-feedback amplifier. http://www.national.com To obtain the closed-loop transfer characteristic, refer again to Fig. 4b. Summing currents at the inverting node yields Ri Vo - Vn (10) since the buffer ensures Vn = Vp = Vi we can rewrite as Vi in Ri II R2 V^ R2 (11) confirming that the feedback signal V0/R2 is now in the form of a current Substituting into Eq. (8), collecting, and solving for the ratio Vo/V, yields A(jf) Vo R2 Ri 1 1/T(jf) (12) where A(jf) represents the c/osed-/oop gain of the circuit, and T(jf) z(if) (13) represents the loop gain. This designation stems again from the fact that if we break the loop as in Fig. 6a, and inject a test voltage Vx with the input V suppressed, the circuit will first convert Vx to the current \„ = — Vx/R2, and then convert In to the voltage Vo = zin, so that T = — (Vo/Vx) = Z/R2, as expected. In an effort to ensure substantial loop gain to reduce the closed-loop gain error, manufacturers strive to make zas large as possible relative to the expected values of R2. Consequently, since L = Vo/z, the inverting-input current will be very small, though this input is a low-impedance node because of the buffer. In the limit z ^ 0° we obtain In -^ 0, indicating that a CF amp will provide whatever output voltage and current are needed to ideally drive In to zero. Thus, the conventional op amp conditions Vn = Vp and In = Ip = hold for CF amps as well. ta(dec) f(dec)-» (b) Fig. 6: Test circuit to find the loop gain, and graphical method to determine the ciosed-loop bandwidth f^. No Gain-Bandwidth Tradeoff The transimpedance gain of a practical CF amp rolls off with frequency according to z(jf) 1 + j(f/fa (14) where Zo is the dc value of the transimpedance gain, and fa is the frequency at which rolloff begins. For instance, from the data sheets of the CLC401 CF amp (Corn- linear Co.) we find Zo = 710 kQ, and fa = 350 kHz. Substituting Eq. (14) into Eq, (13) and then into Eq. (12), and exploiting the fact that R2/Z0 < 1, we obtain A(jf) R2/R1 i(f/fA) (15) where Zofa R2 (16) represents the closed-loop bandwidth, for R2 in the kQ range, fA is typically in the 100 MHz. Retracing previous reasoning, we see that the noise-gain curve is now R2, and that fA can be found graphically as the frequency at which this curve meets the | z | curve, see Fig. 6b. Comparing with Eqs. (6) and (7), we note that the closed-loop gain expressions are formally identical. However, the bandwidth now depends only on R2 rather than on the closed-loop gain 1 + R2/R1. Consequently, we can use R2 to select the bandwidth, and Ri to select the gain. The ability to control gain independently of bandwidth constitutes a major advantage of CF amps over conventional op amps, especially in automatic gain control applications. This important difference is high- lighted in Fig. 7. 1 Gain ( , Coin A=IOO A=IO A=l (a) (b) Fig. 7: Comparing the gain-bandwidth relationship of conven- tional op amps and current-feedback amplifiers. Absence of Slew-Rate Limiting The other major advantage of CF amps is the inherent absence of slew-rate limiting. This stems from the fact that the current available to charge the internal capaci- tance at the onset of a step is proportional to the step regardless of its size. Indeed, applying a step AV induces, by Eq. (11), an initial currentimbalanceln = AVi/(Ri || R2), which the Wilson mirrors then convey to the capacitor. The initial rate of charge is, therefore, Ic/C = L/C = AV/[(Ri||R2)C] = [AVi(1+R2/Ri)]/(R2C) = AVo/(R2C), indicating an exponential output transition with time- constant T = R2C. Like the frequency response, the transient response is governed by R2 alone, regardless of the closed-loop gain. With R2 in the kQ range and C in the pF range, r comes out in the ns range. The rise time is defined as the amount of time tr it takes for the output to swing from 10% to 90% of the step size. For an exponential transition, tr = r X In (0.9/0.1) = 2.2t. http://www.national.com For example, the CLC401 has tr = 2.5 ns for a 2V output step, indicating an effective r of 1 .1 4 ns. The time It takes for the output to settle within 0.1% of the final value is ts = T In 1 000 ^ It. For the CLC401 , this yields ts - 8 ns, in reasonable agreement with the data sheet value of 10 ns. The absence of slew-rate limiting not only allows for faster settling times, but also eliminates slew-rate related nonlinearities such as intermodulation distortion. This makes CF amps attractive in high-quality audio amplifier applications. Second-Order Effects The above analysis indicates that once R2 has been set, the dynamics of the amplifier are unaffected by the closed-loop gain setting. In practice it is found that bandwidth and rise time do vary with gain somewhat, though not as drastically as with conventional op amps. The main cause is the non-zero output impedance of the input buffer, whose effect is to alter the loop gain and, hence, the closed-loop dynamics. Referring to Fig. 8a and denoting this impedance as Ro, we note that the circuit first converts V, to a current Ir2 = Vx/ (R2 + Ri II Ro), then it divides Irs to produce In = Ir2Ri/(Ri + Ro), and finally it converts In to the voltage Vo = zip. Eliminating Ipg and L and letting T = -Vq/Vx yields T = Z/Z2, where Z2 — R2 1 + Ro R1IIR2 (17) Thus, the effect of Ro is to increase the noise gain from R2 to R2[ 1 + Ro/ (Ri IIR2)], see Fig. 8b, curve 1. Conse- quently, both bandwidth and rise time will be reduced by a proportional amount. (a) (b) Fig. 8: Test circuit to investigate the effect of Rq, and noise- gain curves for tlie case of: (1) purely resistive feedbacic, (2) a capacitance in paraiiel witii Rj, and (3) the same capacitance in parallel with R^. Replacing R2 with Z2 in Eq. (16) yields, after simple manipulation. ft 1 -h Ro R2 1 -h R2 \ Ri (18) where ft = zofa/R2 represents the extrapolated value of f A in the limit Ro ^ 0. This equation indicates that bandwidth reduction due to Ro will be more pronounced at high closed-loop gains. As an example, suppose a CF amp has Ro = 50Q, R2 = 1.5 kQ, and ft = 100 MHz, so that fA = 108/[1 -I- (50/1500)Ao] = 108/(1 + Ao/30), where Ao = 1 + R2/R1. Then, the bandwidths corresponding to Ac = 1, 10, and 100 are, respectively, fi = 96.8 MHz, fio = 75.0 MHz, and fioo = 23.1 MHz. Note that these values still compare favorably with a conventional op amp, whose bandwidth would be reduced, respectively, by 1, 10, and 100. If so desired, the external resistance values can be predistorted to compensate for the bandwidth reduction at high gains. Turning Eq. (18) around yields the required value of R2 for a given laandwidth fA and gain Ao. Zofa fA RoAq (19) while the required value of Ri for the given gain Aq is R2 Ri Ao - 1 (20) As an example, suppose we want the above amplifier to retain its 100 MHz bandwidth at a closed-loop gain of 10. Since with R2 = 1 .5 kQ this device has Zofa/R2 = 1 00 MHz, it follows that zofa = 1 0^ X 1 500 = 1 .5 X 1 0" Q X Hz. Then, the above equations yield R2 = 1.5 X lO^VIOs - 50 X 10 = 1 kn, and Ri = 1000/(10- 1) = 111 Q. Besides the dominant pole at fa, the open-loop response of a practical amplifier presents additional poles above the crossover frequency. As shown in Fig. 8b, the effect of these poles is to cause a steeper gain rolloff at this frequency, further reducing the closed-loop bandwidth. Moreover, the additional phase-shift due to these poles decreases the phase margin somewhat, thus causing a small amount of peaking in the frequency response, and ringing in the transient response. Finally, it must be said that the rise time of a practical CF amp does increase with the step size somewhat, due primarily to transistor current gain degradation at high current levels. For instance, the rise time of the CLC401 changes from 2.5 ns to 5 ns as the step size is changed from 2\/ to 5V. In spite of second-order limitations, CF amps still provide superior dynamics. CF Applications Considerations Although the above treatment has focused on the non- inverting configuration, the CF amp will work as well in most other resistive feedback configurations, such as the inverting amplifier, the summing and differencing amplifier, l-V and V-l converters, and KRC active filters." In fact, the derivation of the transfer characteristic of any of these circuits proceeds along the same lines as conventional op amps. Special consideration, however, require the cases in which the feedback network includes reactive elements, either intentional or parasitic. Consider first the effect of a feedback capacitance C2 in parallel with R2 in the basic circuit of Fig. 8a. Letting Z = R2ll(1/sC2), the noise gain becomes Z2 = Z[1 + Ro/(Ri||Z)]. After expanding, it is readily seen that the noise-gain curve has a pole at fp = 1 / (277-R2C2) and a zero at fz = 1/[277-(Ro||Ri ||R2)C2], as shown in Fig. 8b, curve 2. Consequently, the crossover frequency will be pushed into the region of substantial http://www.national.com phase shift due to the higher-order poles of z. If the overall shift reaches —180° at this frequency, then T = 1 /— 180° = —1 there, indicating that A will become infinite by Eq. (12), and the circuit will oscillate. Even if the phase shift fails to reach — 180°, the closed-loop response may still exhibit intolerable peaking and ringing. Hence, capacitive feedback must be avoided with CF amps. To minimize the effect of stray feedback capacitances, man- ufacturers often provide Rz internally. CF Integrators To synthesize the integrator function in CF form, which provides the basis for dual-integrator-loop filters and oscillators as well as other popular circuits, we must use configurations that avoid a direct capacitance between the output and the inverting input. One possibility is offered by the Deboo integrator," which belongs to the class of KRC filters and is therefore amenable to CF realization. Its drawback is the need for tightly matched resistances, if lossless integration is desired. The alterna- tive of Fig. 9 not only meets the given constraint, but also provides active compensation, a highly desirable feature to cope with Q-enhancement problems in dual-integrator- loop filters." Using standard op amp analysis techniques, it is readily seen that the unity-gain frequency of this integrator is fo - (R2/Ri)/(27rRC). reasoning," it is easily seen that f a = [zofafz/ (Ro + F!2) ] ^ '^, where fz = 1/[27r(Ro|| R2)Ci]. Imposing fp = fA yields Ro 277-R2Zofa 1/2 (21) Comp. (a) (b) Fig. 10: DAC output capacitance compensation. To cope with impractically low values of C2, it is con- venient to drive C2 with a voltage divider as in Fig. 10b, since this will scale the value of C2 to the more practical value 1 + Rb ^ (22) It can be shown that for this technique to be effective we must choose Rb < R2. As an example, suppose a DAC having Ci = 1 00 pF feeds the CF amp considered earlier. Then, Eq. (21) yields C2 = [50 X 100 X 10"'^/ (27?- X 1.5 X 103 X 1.5 X 10ii)]i''2 = 1.88 pF To scale it to a more practical value, use ^ Ra = 500 and Rb = 5000. Then, Cc = (1 +500/50) 1.88=: 21 pF This estimate may require some fine tuning to optimize the transient response. Additional useful application hints can be found in Ref. [5]. Fig. 9: Activeiy-compensated CF integrator. Stray Input-Capacitance Compensation Next, consider the effect of an input capacitance Ci in parallel with Ri in the basic circuit of Fig. 8a. Letting Z = Ri||(1/sCi), the noise gain is now Z2 = R2[1 + Ro/(Z||R2)]. After expanding, it is readily seen that the noise-gain curve has a zero at fz = 1/[27r (Roll Ri II R2)Ci], as shown in Fig. 8b, curve 3. If Ci is sufficiently large, the phase of Tat the crossover frequency will again approach —180°, bringing the circuit on the verge of instability. This is of particular concern in current- mode DAC output buffering, where Ci is the output capacitance of the DAC, typically in the range of a few tens to a few hundreds of picofarads, depending on the DAC type. Like a conventional op amp, the CF amp can be stabilized by using a feedback capacitance C2 to introduce sufficient phase-lead to compensate for the phase-lag due to the input capacitance Ci. For a phase margin of 45°, choose the value of C2 so that the noise-gain pole fp = 1 / (277-R2C2) coincides with the crossover frequency fA, as shown in Fig. 10a. Using asymptotic Bode-plot References 1. P. Harold, Current-Feedbacl< Op Amps Ease High- Speed Circuit Design, to be published in EDN, 1988. 2. A New Approach to Op Amp Design, Comlinear Cor- poration Application Note 300-1, March 1985. 3. 1988 Current-Feedbacl< Seminar, Comlinear Corp. 4. Sergio Franco, Design with Operational Amplifiers and Analog ICs, McGraw-Hill Book Company, 1988. 5. Current-Feedback Op Amp Applications Circuit Guide, Comlinear Corporation Application Note OA-07, 1 988. http://www.national.com This page intentionally left blank. http://www.national.com Customer Design Applications Support National Semiconductor is committed to design excellence. For sales, literature and technical support, call the National Semiconductor Customer Response Group at 1-800-272-9959 or fax 1-800-737-7018. Life Support Policy National's products are not authorized for use as critical components in life support devices or systems without the express written approval of the president of National Semiconductor Corporation. As used herein: 1 . Life support devices or systems are devices or systems which, a) are intended for surgical implant into the body, or b) support or sustain life, and whose failure to perform, when properly used in accordance with instructions for use provided in the labeling, can be reasonably expected to result in a significant injury to the user. 2. A critical component is any component of a life support device or system whose failure to perform can be reasonably expected to cause the failure of the life support device or system, or to affect its safety or effectiveness. ^ National Semiconductor Corporation 1111 West Bardin Road Arlington, TX 76017 Tel: 1(800) 272-9959 Fax: 1(800)737-7018 National Semiconductor Europe Fax: (+49)0-180-530 85 86 E-mail: europe.support@nsc.com Deutsch Tel: (+49) 0-180-530 85 85 English Tei: (+49) 0-180-532 78 32 Francais Tel: (+49) 0-180-532 93 58 Italiano Tel: (+49) 0-180-534 16 80 National Semiconductor Hong Kong Ltd. 13th Floor, Straight Blocl< Ocean Centre, 5 Canton Road Tsimshatsui, Kowloon Hong Kong Tel: (852)2737-1600 Fax: (852) 2736-9960 National Semiconductor Japan Ltd. Tel: 81-043-299-2309 Fax:81-043-299-2408 National does not assume any responsibility for use of any circuitry described, no circuit patent licenses are implied and National resen/es the right at any time without notice to change said circuitry and specifications. http://www.national.com Lit #350000-001